Support for Configural Processing in Testing Predictions of Nine Uncanny
Total Page:16
File Type:pdf, Size:1020Kb
Journal of Vision (2021) 21(4):1, 1–20 1 Creepy cats and strange high houses: Support for configural processing in testing predictions of nine uncanny valley theories School of Psychology, Cardiff University, Cardiff, United Kingdom Indiana University School of Informatics and Computing, Alexander Diel Indianapolis, IN, USA Indiana University School of Informatics and Computing, Karl F. MacDorman Indianapolis, IN, USA In 1970, Masahiro Mori proposed the uncanny valley in these markets, valued at more than $270 billion (UV), a region in a human-likeness continuum where an (Research and Markets, 2020). However, human entity risks eliciting a cold, eerie, repellent feeling. characters have also been blamed for box-office flops Recent studies have shown that this feeling can be and studio closures (Freedman, 2012). The effect elicited by entities modeled not only on humans but also identified as disrupting the appreciation of computer nonhuman animals. The perceptual and cognitive animation is the uncanny valley (UV). This term mechanisms underlying the UV effect are not well denotes an observer’s negative affective reaction to understood, although many theories have been human-looking entities, like android robots and proposed to explain them. To test the predictions of nine computer-animated characters. The reaction manifests classes of theories, a within-subjects experiment was as a cold, eerie, repellant feeling. In 1970, Masahiro conducted with 136 participants. The theories’ Mori proposed the UV effect, depicting it with a graph predictions were compared with ratings of 10 classes of (Figure 1; Mori, 2012). stimuli on eeriness and coldness indices. One type of theory, configural processing, predicted eight out of nine Since 2005, UV research has flourished, presenting a significant effects. Atypicality, in its extended form, in wide range of theories. Empirical studies have focused which the uncanny valley effect is amplified by the mainly on determining whether the UV effect exists stimulus appearing more human, also predicted eight. or on testing the predictions of one of these theories Threat avoidance predicted seven; atypicality, (Kätsyri, Förger, Mäkäräinen, & Takala, 2015; Lay, perceptual mismatch, and mismatch+ predicted six; Brace, Pike, & Pollick, 2016; Zhang et al., 2020). More category+, novelty avoidance, mate selection, and than two-thirds of UV studies have found that the psychopathy avoidance predicted five; and category UV effect exists (Burleigh, Schoenherr, & Lacroix, uncertainty predicted three. Empathy’s main prediction 2013; Kim, Bruce, Brown, Visser, & Phillips, 2020; was not supported. Given that the number of significant MacDorman & Ishiguro, 2006; Mathur & Reichling, effects predicted depends partly on our choice of 2015; Mathur et al., 2020; McDonnell, Breidt, & hypotheses, a detailed consideration of each result is Bülthoff, 2012; Seyama & Nagayama, 2007). The effect advised. We do, however, note the methodological value has also been found in infants, children, and nonhuman of examining many competing theories in the same primates (Brink, Gray, & Wellman, 2019; Lewkowicz & experiment. Ghazanfar, 2012; Siebert et al., 2020; Tinwell & Sloan, 2014; Steckenfinger & Ghazanfar, 2009). Nonetheless, its perceptual and cognitive mechanisms are not well understood, nor their neural basis (Rosenthal-von der Introduction Pütten, Krämer, Maderwald, Brand, & Grabenhorst, 2019; Saygin, Chaminade, Ishiguro, Driver, & Frith, The significance of the uncanny valley 2012; Urgen, Kutas, & Saygin, 2018). One way to evaluate theories is to compare the The use of human characters in computer animation, validity of their predictions experimentally across a video games, and special effects has spurred growth range of stimulus conditions. Unfortunately, with only Citation: Diel, A., & MacDorman, K. F. (2021). Creepy cats and strange high houses: Support for configural processing in testing predictions of nine uncanny valley theories. Journal of Vision, 21(4):1, 1–20, https://doi.org/10.1167/jov.21.4.1. https://doi.org/10.1167/jov.21.4.1 Received July 26, 2020; published April 1, 2021 ISSN 1534-7362 Copyright 2021 The Authors Downloaded from jov.arvojournals.org on 08/12/2021This work is licensed under a Creative Commons Attribution 4.0 International License. Journal of Vision (2021) 21(4):1, 1–20 Diel & MacDorman 2 Wolf, & Henze, 2018). Incremental transitions along the dimension are created by morphing (MacDorman & Ishiguro, 2006), editing (Mäkäräinen et al., 2014), or selecting stimuli (Mathur et al., 2020). Stimuli have been derived from photographs of humans, animals, robots, dolls, cartoon and computer-animated characters, and other sources. A few studies used only a small set of stimuli, for example, because the stimuli were difficult to obtain (e.g. physical androids) or because the study investigated the effects of manipulating a specific aspect like facial expressions (e.g. Brink et al., 2019; Palomäki et al., 2018; Tinwell, Nabi, & Charlton, 2013; Złotowski et al., 2015). Mori labeled the y-axis of his graph shinwakan, Figure 1. In 1970, Mori (2012) proposed an N-shaped relation a Japanese neologism indicating social presence between the degree of human likeness of an entity and the and connection, which he translated into English as affinity it elicits in the observer. Affinity increases with human affinity (Mori, 2012). Affinity has been operationalized likeness up to a point before descending into a valley of by one or more scales, typically Likert or semantic eeriness only to rise out again as the entity becomes differential, designed to measure related constructs, indistinguishable from a healthy person. In his graph, the valley such as familiarity, likability, interpersonal warmth,and is steeper when the entity is moving. reverse-scaled eeriness (Bartneck et al., 2009; Ho & MacDorman, 2010, 2017; Redstone, 2013). One aim of UV research is to replicate the UV curve a few exceptions (e.g. MacDorman & Chattopadhyay, using stimuli that vary in human likeness (e.g. Kim et 2016, 2017; Mathur et al., 2020), emphasis has not been al., 2020, Mathur et al., 2020; McDonnell et al., 2012). placed on theory comparison and falsification. Both are Replicating the curve is not our current research goal. needed, given the large number of UV theories that have Another aim is to test differences in the UV effect been proposed. To meet these needs, we have designed between robots or computer-animated characters that and carried out an experiment to test the predictions of vary in some aspect (e.g. Tinwell et al., 2013; Złotowski nine classes of theories prevalent in the UV literature. et al., 2015). Such experiments have already been This investigation is not exhaustive as we have omitted conducted. Instead, this study aims to accomplish what several theories (e.g. the autonomy–heteronomy no other has attempted: to evaluate more than two UV distinction, expectation violation, mortality salience, theories in a single experiment. Given the breadth of and threat to human identity, as discussed in Ferrari, theories considered, the inquiry into each is limited. Paladino, & Jetten, 2016; Gahrn-Andersen, 2020; Follow-up experiments will be needed. MacDorman, Ho, & Vasudevan, 2009b; MacDorman & Ishiguro, 2006; Ramey, 2005; Stein & Ohler, 2017; Urgen et al., 2018). It is also limited to still images, Theories of the uncanny valley although some of the most compelling examples of the UV effect, even predating Mori’s seminal essay, Our review focuses on representative UV theories. involve movement and more than one sensory modality We organized the theories into nine classes based on (Jentsch, 1906; Thepsoonthorn, Ogawa, & Miyake, how their authors presented them; however, two broad 2021). divisions can be made: (1) theories defined in terms of A theoretical understanding of the uncanny valley an entity’s features and their relations versus theories will advance research in related areas, such as social defined in terms of the entity as a whole; and(2) perception, empathy, and human–computer interaction. theories that apply generally versus theories that apply It is also key to identifying design principles for creating in a specific context (Table 1). android robots and computer-animated characters that The term theory has been used in various ways avoid the uncanny valley. because the UV literature is interdisciplinary. For example, although the UV effect has been attributed to cognitive dissonance, this theory is not defined in terms The methodology of uncanny valley research of measurable properties of the stimulus but whether cognitions or actions are consistent. Thus, a degree UV research typically measures affective reactions of interpretation is required to test the predictions of to stimuli that vary in their degree of human likeness. certain theories in the context of vision science. Some studies instead represent the x-axis of Mori’s Configural processing theories predict that the UV graph as perceived realism (e.g. Schwind, Leicht, Jäger, effect is elicited by deviations in the configural pattern Downloaded from jov.arvojournals.org on 08/12/2021 Journal of Vision (2021) 21(4):1, 1–20 Diel & MacDorman 3 Context Features and their relations Entity as a whole Pascalis, & Nelson, 2007; Simpson, Varga, Frick, & Fragaszy, 2011). They found that inconsistency in General Configural processing Category uncertainty feature realism made human and animal faces appear Atypicality Novelty avoidance more unfamiliar and unreal, which elicited cold,